The Worker Retraining (WR) program provides long-term unemployed and dislocated workers with skill training at community and technical colleges.22 Workers must be unemployed or on notice that they are about to be laid off and must be eligible for or have exhausted their unemployment compensation benefits within the last 24 months. The training programs are similar to community and technical college job preparation, i.e., technical training geared to sub-baccalaureate occupations, although funds may also be used for training in basic skills and literacy and related or supplemental instruction for apprentices. Students receive financial assistance to help with tuition and may receive assistance to offset costs of child care and transportation. The trainees are similar in economic circumstances to individuals served by the WIA Title I-B dislocated worker program. In fact, dislocated workers and the long-term unemployed have priority access to the program’s training and supportive services.
Participant Characteristics
Table 7.1 provides descriptive data that compare the individuals in the treatment group to those in the comparison group pool. The comparison group consists of Labor Exchange clients who were 16 to 60 at the time of registration and last received services in 2001/2002 or 2003/2004. The individuals who were served by Washington’s workforce development programs were removed from the comparison group pool data. The first two columns of numbers compare the community and technical college worker retraining clients who exited in 2001/2002 to individuals in the comparison
22A small percentage of Worker Retraining participants attended private career schools, but this project excluded those individuals from the analyses and focused on community and technical college students only.
group. The final two columns compare the community and technical college worker retraining exiters in 2003/2004 to LE exiters in the same year.
The populations were somewhat different. On average, the worker retraining participants are about three years older than the LE exiters. Just under half of the worker retraining clients were females as compared to about 40 percent of the Labor Exchange leavers. Just as was the case with job preparatory training, the worker retraining clients were better educated: a much lower percentage of individuals without a high school diploma (9 percent as compared to over 17 percent for LE registrants in 2001/2002; and 6.6 percent compared to 16 percent in 2003/2004) and higher percentage of individuals with college and with certificates or associate degrees.
Table 7.1 Descriptive Statistics for Worker Retraining Treatment Group and Comparison Group Universe
2001/2002 2003/2004 Characteristics Demographics and Education
Female
College certificate or associate degree Bachelor degree or higher
Limited English proficiency West WA
Employment and Earnings (prior to registration) Ave. percentage of (prior) quarters with employmenta Average quarterly earningsa, b
Mean, earnings trendc
Mean, earnings variancec (in 106 $) Mean, number of quarters with job changeb Had earnings dip
Mean, number of quarters since dip at registrationb Ave. earnings dipb
Sample size 6,048 179,583 7,538 151,842
NOTE: Monetary data in 2000 $.
aObservations with no quarters of prior earnings were excluded from analyses.
bAverages include observations with values of zero.
cTrend and variance calculations include quarters with zero earnings, if any.
††Differences in means are not statistically significant at 0.05 level (t-test).
The Worker Retraining exiters’ work histories showed more employment and higher average quarterly earnings (percentage of quarters worked were about 83 percent versus about 73 for the worker retraining and Labor Exchange clients, respectively). The quarterly earnings difference was over $1,500 in the 2003/2004 data.
Participation Model
Table 7.2 provides the results from the logit estimation of participation. The individuals who had exited from the Labor Exchange were pooled with the community and technical college worker retraining clients, and the dependent variable, participation, was a dummy variable equal to 1 for the latter group (and 0 for the former). The independent variables in the participation model were identical to those used in the job prep participation model documented in Chapter 6. The table provides the logit coefficient estimates and standard errors. While the magnitude of the coefficients is not particularly meaningful, the sign and statistical significance are. If the coefficient is negative, then a (positive) change in that variable will decrease the likelihood of being a community and technical college worker retraining client.
The coefficient estimates seem quite reasonable. The following variables are significantly correlated with being in the treatment group (i.e., a community and technical college worker retraining client) in both years of data: Female, minority, age at registration, having a disability, all the education attainment variables relative to being a high school dropout, being from Western Washington, percentage of quarters employed, turnover (not significant in 2001/2002), having experienced an earnings dip, and magnitude of the earnings dip. The following variables are significantly correlated with being in the comparison group poll: average earnings prior to
registration, variance in earnings prior to registration, and length of time since experiencing an earnings dip.
Propensity Score Statistics
If the participation model had substantial predictive capability, then the mean propensity score for the comparison group should be small (near zero) and should be much less than the mean score for the treatment. As argued earlier, a measure of how well the logit model discriminates between comparison group members and treatment group members is the cumulative percentile for the comparison group at the propensity score that is the 20th percentile. Table 7.3 provides these data for the community college worker retraining analyses. The mean propensity scores for the
Table 7.2 Coefficient Estimates from a Logit Model of Participation in Worker Retraining Program
2001/2002 2003/2004 Characteristics Coefficient Std. Error Coefficient Std. Error
Demographics and Education Female
College certificate or associate degree Bachelor degree or higher
Limited English proficiency West WA Employment and Earnings (prior to registration)
Percentage of quarters with employment Average quarterly earningsa
Earnings trenda Earnings varianceb
Number of quarters with job change Had earnings dip
Number of quarters since dip at registration Percent of earnings dip NOTE:Model included last industry of employment prior to registration and an intercept term. Samples sizes were 185,631 and 159,380 for 2001/2002 and 2003/2004, respectively.
aScaled in $100 (2000 $).
bScaled in $108 (2000 $)
*Significant at the 0.10 level; **significant at the 0.05 level; ***significant at the 0.01 level (two-tailed test).
treatment groups are roughly 0.14 and 0.20, whereas they are 0.03 and 0.04 for the comparison group for 2001/2002 and 2003/2004 respectively. The 20th percentile indicator is approximately 70 percent. The relatively large difference in p-score means is good, but the 20th percentile indicators did not reach the 80th percentile standard.
Table 7.3 Indicators of Propensity Score Model Quality for Worker Retraining Analyses
Statistic 2001/2002 2003/2004
Mean p-score, Worker Retraining 0.140 0.195
Mean p-score, Labor Exchange 0.029 0.040
Percentile Labor Exchange, at 20th percentile Worker
Retraining 66.71% 73.83%
Statistical Match
Table 7.4 provides data about the sample sizes, number of matched observations that were duplicates, and a comparison of descriptive statistics between the treatment group, comparison group, and pool of observations from which the comparison group was chosen. The quality of the match seemed relatively high. Only 10 to 20 percent of the matched comparison group records had multiple copies, and virtually all of the demographic and education as well as employment and earnings means were not significantly different. Relative to the community and technical college job preparatory training, the statistical match for worker retraining did much better on the previous earnings and employment and approximately as well on the educational attainment variables.
Net Impacts
Tables 7.5 and 7.6 provide the estimated net impacts for Worker Retraining. As with comparable tables in the prior chapters, the first table displays the short-term (3 quarters after exit)
and the longer-term (9-12 quarters after exit) outcomes for the 2001/2002 cohort of program exiters.
The second table is limited to the short-term net impacts for the 2003/2004 cohort. The first column in each of the tables presents the estimates from the block matching technique, which uses the full comparison set (i.e., U) and the treatment group. The second column presents a comparison of means between the treatment group and the matched comparison group. The third column presents Table 7.4 Matching Algorithm Statistics and Post-Match Comparison of Characteristics for Worker
Retraining
2001/2002 2003/2004 Statistic/Characteristic
Retraining Labor Exchange Sample size
Sample size used in match Matched sample size
Number of observations used once
Number of observations used multiple times Maximum number of repeats
6,048 Demographics and Education
Female
College certificate or associate degree Bachelor degree or higher
Limited English proficiency West WA
Employment and Earnings (prior to registration) Percentage employed prior to registration Average quarterly earningsa
Mean, earnings trendb
Mean, earnings varianceb (in 106 $) Mean, number of quarters with job changea Had earnings dip
Mean, number of quarters since dip at registrationa Ave. earnings dipa
aAverages include observations with values of zero.
bTrend and variance calculations include quarters with zero earnings, if any.
** Difference in means is statistically significant at the 0.05 level (t-test).
Table 7.5 Net Impact Estimates for Worker Retraining Program for 2001/2002 Cohort Block
Matching Estimator
Matched Sample Estimator
Comparison Group Means
Full Sample Matched Sample Outcome Employment
Short term (%)
Ever employed, longer term (%) Percent of quarters, longer term
Percent of quarters, longer term, diff-in-diff
5.5*** Average hourly wage
Short term ($)
Short term diff-in-diff ($)
0.50***
Longer term, diff-in-diff ($)
0.56*** Average quarterly hours
Short term
Short term, diff-in-diff
29.6*** Longer term
Longer term, diff-in-diff
35.8*** Average quarterly earnings
Short term ($)
Short term, diff-in-diff ($)
284***
Longer term, diff-in-diff ($)
408*** Unemployment Insurance Benefits (average quarterly)
Percent receiving, short term Benefits - short term ($) Percent receiving, longer term
Benefits - longer term ($)
1.7*** Public Assistance and Medicaid (average quarterly)
Percent receiving TANF, short term TANF benefits - short term ($)
0.3
Percent receiving TANF, longer term TANF benefits - longer term ($)
0.4* Percent receiving FS, short term
FS benefits - short term ($) Percent receiving FS, longer term
FS benefits - longer term ($) Percent enrolled Medicaid, short term
Percent enrolled Medicaid, longer term
0.5 NOTE: See Appendix B for explanatory notes. Sample sizes differ for virtually every entry in the table because of observations with missing data. If there were no missing data, the sample sizes would be those displayed in rows 2 and 3 of Table 7.4.
* significant at the 0.10 level; ** significant at the 0.05 level; *** significant at the 0.01 level (two-tailed test). -- means not applicable.
Table 7.6 Net Impact Estimates for Worker Retraining Program for 2003/2004 Cohort Block
Matching Estimator
Matched Sample Estimator
Comparison Group Means
Full Sample Matched Sample Outcome
Average hourly wage Short term ($)
Short term diff-in-diff ($)
0.93*** Average quarterly hours
Short term
Short term, diff-in-diff
43.6*** Average quarterly earnings
Short term ($)
Short term, diff-in-diff ($)
443*** Unemployment Insurance Benefits (average quarterly)
Percent receiving, short term Benefits - short term ($) Public Assistance and Medicaid (average quarterly)
Percent receiving TANF, short term TANF benefits - short term ($)
−0.7*** Percent receiving FS, short term
FS benefits - short term ($) NOTE: See Appendix B for explanatory notes. Sample sizes differ for virtually every entry in the table because of observations with missing data. If there were no missing data, the sample sizes would be those displayed in rows 2 and 3 of Table 7.4.
* significant at the 0.10 level; ** significant at the 0.05 level; *** significant at the 0.01 level (two-tailed test). -- means not applicable.
an estimate from a regression adjustment of that mean. This column represents the preferred specification. For this program we use the levels of the outcome variables as the dependent variable. The coefficient estimates that are in “boxes” represent the final, “official”
estimates using the preferred specification as chosen by WTECB staff. The final columns of the tables provide the means of the comparison group, both the full comparison group pool and the matched comparison group. These columns are provided so that the net impacts can be estimated on a percentage basis.
Short-term impacts for the worker retraining participants, shown in table 7.6, reflect a very strong positive, employment rate gain of 7.8 percentage points and positive net impacts in hourly wages and hours. All together, the estimate of the average change in earnings is about $340. In the short term, the Worker Retraining participants had no significant change in TANF nor Medicaid, and slight reductions in the incidence of unemployment compensation and food stamps.
The longer-term earnings impacts were similar, although smaller in magnitude. The employment rate and hours of employment increased at levels that were statistically significant (4.4 percentage points and 29.8 hours, respectively), but the increase in the average hourly wage rate (0.16) was not significant. All together, the earnings impact was about $270. In addition, the net impacts on public assistance and unemployment compensation showed increases that were statistically significant.
Subgroup Analyses
About half of the 2003/2004 treatment group and about 40 percent of the 2001/2002 Worker Retraining treatment group actually completed their community and technical college course of study. Selected net impact estimates for these subgroups are provided in tables 7.7 and 7.8. As with the job prep students analyzed in the previous chapter, in both the short term and longer term, the completers have more positive outcomes. All of the employment and earnings impacts for completers are positive, significant, and larger than the impacts for noncompleters. In fact, table 7.7 shows that the hourly wage impact for noncompleters is negative (although not significant); and consequently, the earnings impact is virtually 0. The public assistance and UI impacts tend to be negative (i.e., reductions in take-up) for completers as compared to less negative or positive impacts for the noncompleters.
Table 7.7 Selected Longer Term Net Impact Estimates for Subgroups of Worker Retraining Participants:
2001/2002 Cohort
Subgroup
Outcome Noncompleters Completers
Matched Comparison Group Mean
Employment 2.6%** 8.2%** 59.7%
Hourly Wage −$0.29 $0.99** $9.90
Hours Worked 15.6** 53.7** 246.6
Earnings $12 $712** $3,947
UI Receipt 1.0% 2.3%** 13.4%
TANF Receipt 0.7%** −0.1% 2.8%
Food Stamps Recipient 3.1%** −0.6% 11.2%
Medicaid Enrollment 2.7%** −1.3%** 11.2%
Subgroup Sample Size 3,711 2,320 —
NOTE: Monetary data in ’00 $.
**Significant at the 0.05 level (two-tailed test).
Table 7.8 Selected Short-Term Net Impact Estimates for Subgroups of Worker Retraining Participants:
2003/2004 Cohort
Subgroup
Outcome Noncompleters Completers
Matched Comparison Group Mean
Employment 4.6%** 10.8%** 61.7%
Hourly Wage $0.59** $0.83** $9.46
Hours Worked 20.5** 57.8** 241.9
Earnings $174 $520** $3,664
UI Receipt −0.2% −1.7%** 6.2%
TANF Receipt −0.1% −0.5%** 3.6%
Food Stamps Receipt −0.0% −1.7%** 13.9%
Medicaid Enrollment 1.2%** −0.9% 14.4%
Subgroup Sample Size 3,646 3,873 —
NOTE: Monetary data in ’00 $.
**Significant at the 0.05 level (two-tailed test).
8 ADULT BASIC EDUCATION PROGRAMS ON COMMUNITY AND